%0 Conference Paper %1 705d342cff7449e7bdb2f77a45ca0c0b %A Incardona, Pietro %A Bianucci, Tommaso %A Sbalzarini, Ivo F. %B High Performance Computing %C Germany %D 2021 %E Chamberlain, Bradford L. %E Varbanescu, Ana-Lucia %E Ltaief, Hatem %E Luszczek, Piotr %I Springer, Berlin u. a. %K Block CUDA, Distributed FIS_scads GPU, Sparse data, grid grid, %P 272--290 %R 10.1007/978-3-030-78713-4_15 %T Distributed Sparse Block Grids on GPUs %X We present a design and implementation of distributed sparse block grids that transparently scale from a single CPU to multi-GPU clusters. We support dynamic sparse grids as, e.g., occur in computer graphics with complex deforming geometries and in multi-resolution numerical simulations. We present the data structures and algorithms of our approach, focusing on the optimizations required to render them computationally efficient on CPUs and GPUs alike. We provide a scalable implementation in the OpenFPM software library for HPC. We benchmark our implementation on up to 16 Nvidia GTX 1080 GPUs and up to 64 Nvidia A100 GPUs showing state-of-the-art scalability (68% to 96% parallel efficiency) on three benchmark problems. On a single GPU, our implementation is 14 to 140-fold faster than on a multi-core CPU. %@ 9783030787127

@inproceedings{705d342cff7449e7bdb2f77a45ca0c0b, abstract = {We present a design and implementation of distributed sparse block grids that transparently scale from a single CPU to multi-GPU clusters. We support dynamic sparse grids as, e.g., occur in computer graphics with complex deforming geometries and in multi-resolution numerical simulations. We present the data structures and algorithms of our approach, focusing on the optimizations required to render them computationally efficient on CPUs and GPUs alike. We provide a scalable implementation in the OpenFPM software library for HPC. We benchmark our implementation on up to 16 Nvidia GTX 1080 GPUs and up to 64 Nvidia A100 GPUs showing state-of-the-art scalability (68% to 96% parallel efficiency) on three benchmark problems. On a single GPU, our implementation is 14 to 140-fold faster than on a multi-core CPU.}, added-at = {2024-11-28T16:27:18.000+0100}, address = {Germany}, author = {Incardona, Pietro and Bianucci, Tommaso and Sbalzarini, {Ivo F.}}, biburl = {https://puma.scadsai.uni-leipzig.de/bibtex/2d5273463f58ba6040b9d051381bb7cb2/scadsfct}, booktitle = {High Performance Computing}, doi = {10.1007/978-3-030-78713-4_15}, editor = {Chamberlain, {Bradford L.} and Varbanescu, Ana-Lucia and Ltaief, Hatem and Luszczek, Piotr}, interhash = {de84ba0851486efe8920b674ca5f1fdd}, intrahash = {d5273463f58ba6040b9d051381bb7cb2}, isbn = {9783030787127}, keywords = {Block CUDA, Distributed FIS_scads GPU, Sparse data, grid grid,}, language = {English}, note = {Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 36th International Conference on High Performance Computing, ISC High Performance 2021 ; Conference date: 24-06-2021 Through 02-07-2021}, pages = {272--290}, publisher = {Springer, Berlin [u. a.]}, series = {Lecture Notes in Computer Science, Volume 12728}, timestamp = {2024-11-28T16:27:18.000+0100}, title = {Distributed Sparse Block Grids on GPUs}, year = 2021 }